Vehicle-to-Vehicle Safety Messaging in DSRC

Vehicle-to-Vehicle Safety Messaging in DSRC Qing Xu * Raja Sengupta * Department of Mechanical Engineering Department of Civil and Environmental Engineering University of California Berkeley, CA 90720 1740 Email: * qingxu@me.berkeley.edu raja@path.berkeley.edu Tony Mak Jeff Ko California Partners of Advanced Transits and Highways (PATH) Richmond, CA 94804-4603 Email: {tonykm, jko}@path.berkeley.edu Abstract This paper studies the design of layer-2 protocols for a vehicle to send safety messages to other vehicles. The target is to send vehicle safety messages with high reliability and low delay. The communication is one-to-many, local, and geo-significant. The vehicular communication network is ad-hoc, highly mobile, and with large numbers of contending nodes. We design several random access protocols for medium access control. The protocols fit in the DSRC multi-channel architecture. Analytical bounds on performance of the addressed protocols are derived. Simulations are conducted to assess the reception reliability and channel usage of the protocols. The sensitivity of the protocol performance is evaluated under various offered traffic and vehicular traffic flows. The results show our approach is feasible for vehicle safety messages in DSRC. I. INTRODUCTION Dedicated Short-Range Communications (DSRC) is 75 MHz of spectrum at 5.9 GHz allocated by the Federal Communications Commission (FCC) to increase traveller safety, reduce fuel consumption and pollution, and continue to advance the nation s economy [7]. It is a promising development supporting vehicle to vehicle and vehicle to infrastructure communication using a variant of the IEEE 802.11a technology [1]. DSRC would support safety-critical communications for say collision warnings, as well as other valuable ITS applications such as Electronic Toll Collection (ETC), digital map update, etc. The versatility of DSRC greatly enhances the likelihood of its deployment by various industries and adaptation by the consumers. The 2004 FCC ruling [8] specifies DSRC will have six service channels and one control channel. The control channel is to be regularly monitored by all vehicles. Safety messages, whether originated by vehicles or roadside transmitters, are to have priority over non-public safety communications. Therefore all safety messages are to be sent in the control channel. In the meantime, a licensed roadside unit could use the control channel to inform approaching vehicles of its services (often nonsafety applications) and conduct the actual application in one of the service channels. For example, a roadside unit could announce a local digital map update in the control channel and transfer this data to interested vehicles in a service channel. Table I illustrates the sort of DSRC data traffic characteristics being used in the standards deliberations [6] [12] [19]. The FCC has recognized safety messages and safety of life messages. Safety of life is to have the highest priority. The non-safety data transfers have the lowest priority. The non-safety communications happen for file transfers (e.g., infotainment) or transactions (e.g. toll collection). Transactions may require the exchange of two or three messages within a short time window (e.g., 20 msec). This paper explores the feasibility of sending safety messages from vehicle to vehicle in the DSRC control channel. Safety messages are time sensitive. When vehicles send safety messages to each other while travelling at high speed, can they be received with small delay and high probability? Cellular networks achieve time sensitive communication at high speeds. But they do so with the aid of base stations. These are significantly more expensive than their supposed DSRC equivalent, i.e., 802.11 access points. Moreover, cellular handles only infrastructure to mobile communication. Can DSRC handle time and loss sensitive vehicle-vehicle communication? Can it do so without infrastructure, i.e., in ad hoc setting? Will there be enough room on the control channel for the other non-safety communications required by DSRC, e.g., service announcements? After all, if the announcements are crowded out the remaining

TABLE I TYPICAL DSRC DATA TRAFFIC REQUIREMENTS TABLE II OFFERED TRAFFIC PARAMETER RANGES Packet Allowable Network Message Applications Size(bytes) Latency Traffic Range(m) Priority /Bandwidth (ms) Type Intersection Collision 100 100 Event 300 Safety Warning/ of Life Avoidance Cooperative Collision 100/ 100 Periodic 50-300 Safety Warning 10Kbps of Life Work Zone 100/ 1000 Periodic 300 Safety Warning 1Kbps Transit Vehicle 100 1000 Event 300 - Safety Signal Priority 1000 Toll Collection 100 50 Event 15 Non -Safety Service Announce- 100/ 500 Periodic 0-90 Non ments 2Kbps -Safety Movie Download (2 hours >20 Mbps N/A N/A 0-90 Non of MPEG 1): -Safety 10 min. download time six channels will become unusable. The rest of the paper is structured as follows. Section II reviews the relevant literature. Section III is the problem formulation, IV discusses the protocol design, V explains our evaluation methods, and VI shows the evaluation results. Section VII summarizes the main findings. It answers the questions in the introduction. II. LITERATURE REVIEW In ad hoc vehicular networks, TDMA, FDMA, or CDMA are difficult due the need to dynamically allocate slots, codes, or channels without centralized control. We base our designs on random access [17]. ALOHA [3] and CSMA [17] are the earliest studied random access protocols. MACA [11], MACAW [5], FAMA and its variants [9] all use the RTS/CTS scheme. Our communication is broadcast (see Section III). We do not use RTS/CTS. HIPERLAN/1 [4], Black Burst [16], and the Enhanced Distributed Coordination Function (EDCF) of IEEE 802.11e [20] are all designed to support QoS. The HIPERLAN/1 and Black Burst approaches have no scheme to combat hidden terminals. In EDCF, when the number of contending packets of equal priority is large the probability of collision is still high. This is the case for vehicle safety communications (Section III). Reference [15] reviews the existing variants of the 802.11 DCF to support QoS. Its authors conclude that the design of a mechanism to provide predictable QoS in an 802.11 network is still an open problem. We use a Message Generation Interval (msec) 50, 100, 200 Packet Payload Size (Bytes) 100, 250, 400 Data Rate (Mbps) 6, 9, 12, 18, 24, 36, 48, 54 Average Vehicle Distance (m) 10 (jammed) 30 (smooth) Message Range (m) 10-100 30-300 Lane Number 4, 8 different definition of QoS (Section III). Reference [22] gives an overview of DSRC applications and assesses the characteristics of the IEEE 802.11 MAC and PHY layers in this context. It is anticipated that the current 802.11 specifications will need to be suitably altered to meet the QoS requirements of DSRC applications. III. PROBLEM FORMULATION This paper evaluates the feasibility of sending vehiclevehicle and roadside-vehicle safety messages. These messages should be received reliably and with small delays. We define the Probability of Reception Failure (PRF) as the probability a targeted receiver fails to receive a safety message within a given time delay. The safety messages are to be sent on the control channel. The control channel also has to communicate other non-safety messages for the remaining channels to be used. Therefore the fraction of the control channel time occupied by safety messages is important as well. This is measured by channel busy time (CBT). Both PRF and CBT should be low. To evaluate feasibility we have had to make an assessment of the offered safety traffic. When the offered traffic is large, reliability, latency, and CBT deteriorate. In wired networks offered traffic is measured by the total bits/second produced by all the senders. In wireless networks the right measure of offered traffic is bitmeters/second [10], i.e., a network able to transmit a bit 100 meters, may not be able to transmit the same bit 200 meters. Therefore the offered traffic depends on the safety message rate (messages/sec), size (bytes/message), message range (meters), and the density of vehicles producing these messages. Table II gives ranges for the parameters determining the offered traffic. Our evaluation is based on these ranges. A vehicle at high freeway speeds (90 mph) moves 2 meters within its lane in 50 msec. This is usually not a significant movement at high speed. Thus messages repeating faster than once every 50 msec are unlikely to provide significantly new information. On the other hand

an update slower than once every 500 msec is probably too slow. Driver reaction time to stimuli like brake lights can be of the order of 0.7 seconds and higher. Thus if updates come in slower than every 500 msec, the driver may realize something is wrong before the safety system. Message sizes have been chosen to permit sender or receiver location as per the SAE J1746 standard, GPS, NTCIP hazard codes, and standard protocol headers to be included. Safety messages are usually short. Communication is more difficult at high vehicle densities. The 10 meters per vehicle represents the jammed highway. The 30 meters per vehicle represents the highway at capacity. Likewise, the 4 to 8 lane range spans the usual to large roads. Safety application designers would prefer large message ranges to smaller ones. On the other hand large message ranges make network design more difficult. The 300-meter message range corresponds to the comfortable stopping distance of a high speed car. When the road is jammed, neighboring cars will be much closer. Therefore it should not be necessary to send safety messages over the same distance. We assume a top range of 100 meters for jammed roadways, or approximately 10 inter-vehicle distances. We have proposed a communication service able to execute at least the vehicle-vehicle communication without any roadside or base station infrastructure, i.e, an ad-hoc service. This would be good for deployment. Since 802.11a radios are designed to transmit over distances of 200 to 300 meters, i.e., the upper end of the message range in Table II, we propose a single hop, local area communications service. Reliable communication in networks has typically meant re-transmitting a message till it is acknowledged by the recipient(s). This is good for file transfers since even one missing byte may render the entire file unusable. Thus reliable transmission protocols like TCP ensure each byte is received with certainty. For DSRC safety messages we have proposed a local area communication service that only delivers messages with high probability. This is for several reasons. We think most safety messages, like those in Table I, will be repeated by the source. For example, a broken down vehicle stopped or moving slowly in a high-speed arterial, would transmit its status, position, and speed repeatedly. This is because the set of vehicles approaching the stopped vehicle changes. If the message repeats every 100 msec and a message has still not been received 100 msec after it was created, the source will produce a new message obsoleting the old one. The communication service should also drop the old one and work on the new one. We think most safety messages have a useful lifetime. Therefore we have focused on the design of a single-hop local-area communication service delivering messages within their lifetime with high probability. The lifetime is the delay requirement. Therefore the PRF is a targeted receiver s probability of failing to receive a message within its lifetime. We think of the lifetime and probabilityof reception failure as the Quality of Service (QoS) requirement of the message. Our evaluation focuses on senders that generate periodic or Poisson distributed messages. If the active safety systems on the vehicle assist the driver rather than substitute for her, we think probabilities of reception failure of 1/1000 to 1/100 per message may be adequate. In any case, most safety messages should be consumed by an estimator. For example, the warnings from the slowly moving or stopped vehicle should be consumed by an estimator of the position of the damaged vehicle relative to the receiver, conditioned on all received messages and possibly sensor information as well. Since an estimator leverages correlations in the time series of messages, it is usually robust to the loss of messages, unless the losses occur in bursts. Most safety messages produced by a vehicle are useful to many vehicles. For example, the stopped vehicle warning is useful to all approaching vehicles. Therefore we have focused on a broadcast service. In summary, we propose a service to broadcast messages while meeting QoS requirements in vehicular adhoc local-area networks. IV. PROTOCOL DESIGN In a wireless ad-hoc network there are two obstacles to the reliable reception of messages. If two transmitters within the interference range of a same receiver transmit concurrently, their transmissions collide at the receiver. The receiver does not receive either message. To combat this problem one designs a Medium Access Control (MAC) protocol, i.e., a set of rules by which a radio decides when to transmit its messages and when to keep silent. Secondly, even if there is no collision, the wireless channel may attenuate the transmitted power so much that it is swamped by thermal noise. This is combated by selecting the transmission energy to be high enough to reach all receivers within the message range with high probability, when there are no collisions. Transmission energy is determined by transmission power, modulation, and error coding. DSRC radios are to be based on the 802.11a radio. In our evaluation we

set the transmission energy control parameters to model the 802.11a radio transmitting over a 20 MHz channel at 5.4 GHz and focus on the MAC design problem, i.e., is there a MAC able to deliver safety messages with sufficiently high reliability and small delays? The stochastic modelling of the wireless vehicle-to-vehicle communication channel is an open problem. We use the deterministic Friis Free-space model for short distances and the Two-ray model for longer distances [13], i.e, if the distance between the transmitter and receiver antennas is d, then the power of the signal decreases as d 2 when the distance is short and d 4 when the distance is large. In unicast communication reliability is enhanced by policies based on receiver feedback, e.g. RTS/CTS, TCP, or WTP. These require the sender to learn the identity of its receiver(s). When there are many receivers or the network is highly mobile, meaning the set of receivers can change a lot, learning identities may itself require significant communication. Therefore we have chosen to evaluate ways to enhance reliability without receiver feedback. Our strategies repeat each message without acknowledgement in combination with CSMA and its variants. Our repetition schemes are designed for overlay on CSMA. The following is the specifications of our various designs. A. Protocol Specifications Figure 1 is an illustration of the idea of repetitive transmission. It shows two transmitters within interference range of one receiver each generating a message at the same time. Every repetition of the message is a new packet. At each transmitter the protocol evenly divides the message lifetime into n = τ t trans slots, where x is the maximum integer not greater than x, τ is the lifetime, and t trans is the time needed to transmit one packet. We randomly pick any k (1 k n) slots to repetitively transmit the message. If any one or more of the packets corresponding to the message are received without collision at a given receiver, the message is received within its useful lifetime. On the other hand, the message fails if all of its transmitted packets are lost due to collisions. Our protocols use two schemes to reduce PRF, repetition and carrier sensing. Carrier sensing is in the 802.11 MAC. In all the cases but two, our protocol is an overlay on the standard MAC like ALOHA [3] or Carrier Sensing [17]. We call this overlay the MAC extension layer. Message 1 Generated Message 2 Generated Fig. 1. Fig. 2. Useful lifetime The Concept of Repetitive Transmission MAC Extension Layer State Machine Our MAC extension layer would lie between the Logical Link Control layer (IEEE 802.2) and the standard MAC layer. Its role is to generate and remove repetitions. The state machine of the MAC extension layer is shown in Figure 2. Upon receiving a message from LLC, MAC Extension transits from IDLE to REPETITION GENERATION state. In this state, the system schedules multiple repetitions of this message in the selected time slots within the message lifetime. Each repetition is an event with a slot number. All these events ordered by slot numbers form a queue called the Packet Event Queue. Once the queue is formed, the system transits back to the IDLE state. Whenever a packet event expires, the MAC extension transits to the DISPATCH state and sends the packet down to the MAC. The system then transits back to IDLE. Whenever MAC Extension receives a packet from MAC, the system transits from IDLE to REPETITION REMOVAL state. If the message ID in this packet has not been seen before, it is from a new message, and the new message is passed up to the LLC. If the message ID in this packet has been seen before, the packet is eliminated. The following are different protocols designed and

evaluated by us. They share the same MAC extension layer. 1) Asynchronous Fixed Repetition (AFR) AFR is configured by setting the number of repetitions k. The protocol randomly selects k distinct slots among the total n slots in the lifetime. The protocol is called fixed because the packet is always repeated a fixed number of times, i.e., k. The radio does not listen to the channel before it sends a packet with AFR. 2) Asynchronous p-persistent Repetition (APR) The p-persistent repetition protocol determines whether to transmit a packet in each of the n slots in the lifetime by flipping an independent unfair coin with P (H) = k n and P (T ) = 1 k n. A packet is transmitted if the result is head, where k is again a configuration parameter of the protocol. We can see the average number of transmissions of a message will be k. However, for each realization the exact number of repetitions varies. Like AFR, the radio does not listen to the channel before it sends a packet. 3) Synchronous Fixed Repetition (SFR) This protocol is the same as AFR except that all the slots in all the nodes are synchronized to a global clock like slotted ALOHA [3]. 4) Synchronous p-persistent Repetition (SPR) The SPR protocol is the same as the APR protocol except for the synchronization of transmissions by all nodes into common slots. 5) Asynchronous Fixed Repetition with Carrier Sensing (AFR-CS) AFR-CS has its own MAC shown in Figure 3. AFR-CS generates the repetitions in the same way as in the AFR protocol. Whenever a packet is passed down from the MAC Extension, MAC transits from the IDLE to the CARRIER SENS- ING state. In the CARRIER SENSING state, the system checks the channel status using carrier sensing [17]. If the channel is busy, the system drops the packet and transits back to the MAC IDLE state. If the channel is idle, the system transits to the MAC TX state, and passes the packet down to the physical layer (PHY). It then transits back to the MAC IDLE state. In MAC IDLE, if PHY sends a packet up, the system transits to the MAC RX state and checks the integrity of the packet. If the packet is corrupted, it is dropped and the system transits back to the MAC IDLE state. Fig. 3. MAC Layer State Machine of the AFR-CS protocol Otherwise, the packet is passed up to the MAC Extension layer, and the system transits back to the MAC IDLE state. 6) Asynchronous p-persistent Repetition with Carrier Sensing (APR-CS) This is similar to AFR-CS except that the slots for message repetitions are selected in the p-persistent manner. V. EVALUATION METHOD We have two methods of evaluation. For the SPR and APR protocols we have developed mathematical expressions for probability of reception failure. These expressions can be processed using Matlab. Secondly we have developed a DSRC simulator. The simulator is based on two others, namely SHIFT [14] and NS-2 [2]. SHIFT is a well established traffic simulator. It gives us the trajectories of vehicles driving according to validated models on realistic road networks [18]. Figure 4 is a screenshot of the vehicle traffic simulated by SHIFT. We use SHIFT to generate the motion of the radios. This motion is input to NS-2. NS-2 is an open source network simulator widely used in academic community. NS-2 generates the offered traffic, simulates transmissions and receptions, and outputs packet reception data. We postprocess the data to obtain channel busy time, probability of reception failure, and the probability of long bursts of failures. Thus the DSRC simulator is the standard NS-2 release plus SHIFT The radio model for 802.11a at 5.4 GHz The repetition protocols A different data structure that changes the run-time of NS-2 from quadratic to linear in the number of

Fig. 4. A typical traffic screen-shot of SHIFT nodes. Figure 5 shows the run-time comparisons. This enhancement has enabled us to simulate networks with up to a thousand vehicles. 1200 upper and lower bounds of the PRF for SPR at a receiver with m interferers satisfies the following inequality ( 1 k ) n n e mλτ k n < ( P ( S) < 1 k ) k n n e mλτ k n + n e mλτ (1) With APR, the PRF at a receiver with m interferers satisfies the following inequality. ( 1 k n e mλτ[ 2 k k 2 ]) n n n 2 < ( P ( S) < 1 k n e mλτ[ 2 k k 2 ] n n 2 + k ) n n e mλτ (2) Simulation time (sec) 1000 800 600 400 200 PATH DSRC Simulator NS 2 Mobile Nodes NS 2 Stationary Nodes 0 100 200 300 400 500 600 700 800 900 1000 Number of nodes Fig. 5. Improvement on the Scalability of PATH DSRC Simulator over NS-2 We have used the Matlab expressions to check that the simulator is free of software errors. The Matlab expressions being computationally simpler, we use them to get a qualitative understanding of what the simulator might show. The parameter space to be explored is large. The mathematical expressions have also been used to select parts of the parameter space for intensive simulation. The exploration of the parameter ranges in Table II has produced 400 GB of archived data. Both simulator and data are available to others. VI. EVALUATION RESULTS We study homogeneous system in this paper. The repetition number k, transmission power, data rate, and packet size are the same for all nodes. In the mathematical analysis we assume the inter-vehicle distances are the same. In the simulation we specify average flow of the highway, but the the inter-vehicle distances vary. A safety message is successful if each receiver within its message range receives at least one of the repetitions. When the message generation processes are Poisson, In the inequalities above, n is the total number of slots, i.e. the maximum possible number of repetitions in the message lifetime, k is the number of repetitions for the message (average value for p-persistent protocols and exact value for fixed repetition protocols), S stands for the event that at least one of the repetitions succeeds, τ is the lifetime of the message, λ is the message generation rate at each individual node, and m represents the total number of interferers around a receiver. The details of the mathematical analysis are in [21]. The number of interferers m in equations (1) and (2) is calculated by: 2 Interference Range Interferer Number = Meters per Vehicle Lane number (3) The procedure to calculate the interference range r i, given the distance between transmitter and receiver r, the message range R, and the date rate, is as follows. We need to first calculate the transmission power P t required of R, and then the interference range r i for r. There is a Signal to Interference+Noise Ratio (SINR) threshold for a radio to receive data at a given data rate. The higher the data rate, the higher the threshold. Specific values of the thresholds depend on the radio design. In our simulation we use the SINR threshold values of a commercial off-the-self 802.11a radio product. The procedure to calculate transmission power P t of a message to reach a message range R at a given data rate is the following. 1) Find the SINR threshold β corresponding to the data rate. The ratio between the reception power P r and thermal noise N is β in db. 2) Calculate the reception power P r = N 10 β 10. 3) Calculate the transmission power

Use the path-loss channel model, P r, and R to calculate P t. If the free-space model is used, P t = P r R2 A, with A being some constant depending on the signal wavelength and the gains of transmit and receive antennas. Given the transmission power P t, the TX/RX distance r R, and the data rate, the procedure to calculate the interference range r i of the receiver is the following. 1) Find the SINR threshold β corresponding to the data rate. 2) Calculate the power of the signal at the receiver P r If the free-space model is used, P r = P t A r. 2 3) Calculate the minimum power required to interfere The ratio between P r and the interference power P i is equal to or lower than β in db, hence, P i = 10 β 10 P r. 4) Calculate r i Use the channel model, P t, and P i to calculate r i. If the free-space mode is used, r i = APi P t. All nodes closer than r i from the receiver can interfere. The interference range depends on the TX/RX distance. If free-space channel model is used, we have the following relation. r i = 10 β 20 r (4) Hence, the interference range as well as the number of interferers of a receiver is a linearly increasing function of its distance to the transmitter. All the following results were taken when the distance between the transmitter and receiver is the message range. So all the results are for the worst case. Intuition suggests when a message is repeated more than once the chances of it being received may rise. On the other hand repetition increases the aggregate number of collisions, implying repetition beyond a certain level should be counter-productive. Figure 6 is a Matlab plot of the equation (1) for the parameters values in Table III. It confirms this intuition. There is an optimal number of repetitions. This optimal number is different for different message ranges, message generation rates, vehicular traffic densities, message sizes, etc. Figure 7 shows the variation of the optimal repetition number and the minimum PRF with message range. Other parameter values are as in Table III. Figure 8 shows the performance of the protocols as a function of the number of repetitions. The curves are based on output from the DSRC simulator. CSMA Probability of Reception Failure 10 0 10 1 10 2 0 10 20 30 40 50 60 Number of Transmissions Fig. 6. Probability of Reception Failure vs. Number of repetition: SPR in the nominal setting, analytical PRF generated by Matlab Probability of Reception Failure 10 0 10 1 10 2 10 3 10 4 10 5 0 10 20 30 40 50 60 70 80 Number of Transmissions 30 m 80 m 110 m Fig. 7. Probability of Reception Failure at different message ranges: SPR in the nominal setting, analytical PRF generated by Matlab parameters (e.g., DIFS) have been kept fixed at their 802.11 values. The best protocols are AFR-CS and SFR. But SFR would require a clock synchronization infrastructure. Therefore AFR-CS seems to be the best solution. Note also the PRF improvement over 802.11a is one order of magnitude. Figure 9 is a plot of CBT versus repetition number for the two protocols. It shows that CBT increases with the repetition number. With the same number of repetitions the AFR-CS protocol has smaller CBT than other proposed protocols. Not surprisingly, the 802.11 has small CBT since it does not repeat. These plots are for the nominal parameters in Table III. Therefore the rest of our evaluation uses AFR-CS. The Channel Busy Time is a measure of the fraction of channel capacity left over for non-safety messages. As the repetition number goes up so should CBT. Thus there is an inverse relationship between PRF and CBT

Probability of Reception Failure 10 0 10 1 10 2 Asynchronous fixed repetition, 18 Mbps Fixed repetition CSMA, 18 Mbps Asynchronous p persistent, 18 Mbps P persistent CSMA, 18 Mbps Synchronous fixed repetition, 12 Mbps Synchronous p persistent, 12 Mbps 802.11, 24 Mbps TABLE III NOMINAL SETTING PARAMETERS Message Generation Interval (msec) 100 Useful Life Time (msec) 100 Packet Payload Size (Bytes) 100 Message Range (m) 80 Average Distance Between Vehicles (m) 30 Lane Number 4 10 3 10 4 0 2 4 6 8 10 12 14 16 18 20 Number of Transmissions Fig. 8. Probability of Reception Failure for Proposed Protocols in the Nominal Setting 0.8 0.7 0.6 Fixed Repetition with CSMA Asynchronous Fixed Repetition Sychronous Fixed Repetition 802.11 Probability of Reception Failure 10 0 10 1 10 2 10 3 Headway 10 m, Msg. Range 25 m, 4 Lanes, 70 interferers Headway 30 m, Msg. Range 80 m, 4 Lanes, 75 interferers Headway 30 m, Msg. Range 120 m, 4 Lanes, 113 interferers Headway 30 m, Msg. Range 60 m, 8 Lanes, 113 interferers Headway 20 m, Msg. Range 80 m, 4 Lanes, 113 interferers Headway 30 m, Msg. Range 160 m, 4 Lanes, 151 interferers Headway 30 m, Msg. Range 80 m, 8 Lanes, 151 interferers Headway 15 m, Msg. Range 80 m, 4 Lanes, 151 interferers Channel Busy Time 0.5 0.4 0.3 0.2 10 4 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Channel Busy Time Fig. 10. Performance of AFR-CS Protocol as a Function of Interference Indicator 0.1 0 0 2 4 6 8 10 12 14 16 18 20 Fig. 9. Number of Transmissions Channel Busy Time in the Nominal Setting up to the optimal repetition number. Curves in Figure 10 show this tradeoff. As the quality of service to the safety applications is raised, that of for other applications in control channel comes down. Therefore these curves are a good way of evaluating the performance of our protocols. Equations (1) and (2) shows performance depends on the number of interferers. In an actual traffic simulation this number varies in complex patterns. Nevertheless the single number evaluated by (3) is a good predictor of the performance. Figure 10 shows that for small transmission time the dynamics of the topology makes almost no difference to aggregate network performance. The curve for 30-meter headway, 60-meter message range, 8 lanes and the curve for 20-meter headway, 80-meter message range, and 4 lanes are clustered together because they have the same number of interferers calculated by (3), i.e., 113. The number of interferers for the nominal case is 75. Intuition suggests that for each number of interferers and repetition number there may be an optimal data rate. If the transmission rate is raised, the transmission time reduces, tending to reduce the probability of collision. On the other hand the power required to cover the message range also rises, thereby raising the number of interferers. Figure 11 shows this for AFR-CS protocol with the nominal parameter combination in Table III. Table IV shows the optimal data rates for all the protocols in the nominal parameter setting. Figure 12 shows the probabilities of bursts of fail- TABLE IV OPTIMAL DATA RATE FOR ALL THE SIMULATED PROTOCOLS IN THE NOMINAL SETTING Protocol Optimal Data Rate (Mbps) SFR 12 AFR 18 SPR 12 APR 18 APR-CS 18 AFR-CS 18 802.11 24

7 x 10 3 80m Communication Range Asymptotic Probability of Reception Failure 6 5 4 3 2 Probability of Occurence 0.08 0.06 0.04 1 0.02 0 6 9 12 18 24 36 48 54 Data Rate (Mbps) Fig. 11. Probability of Reception Failure for Various Data Rate in the Nominal Setting: AFR-CS Protocol 0 1 2 4 8 11 Number of Repetitions 15 20 6 5 1 2 3 4 Number of Consecutive Failures ures of different lengths for the nominal parameters in Table III. As expected the probabilities are small. The probability for a receiver seeing two or more consecutive failures is negligible. Figure 13 pulls these various results together to get a sense of the feasibility of supporting safety applications on the control channel. Feasibility depends on a PRF and CBT requirement. Given a PRF and CBT requirement and a combination of parameter values within the ranges in Table II, our simulation data can show whether it meets the requirements. We do this by assuming the AFR-CS protocol and optimizing the protocol for repetition number, transmission rate, and selecting the modulation and code rate to minimize the power required to cover the message range. Figure 13 shows feasibility for a PRF less than 1/100 and a CBT less than 50%. For example, the 200 msec message rate, 250 byte message at 140 interferers is feasible. This corresponds to a30 meter, 4 lane highway at capacity (2200 vehicles/hour/lane) with a message range of 150 meters. Likewise the 10 meter headway (jammed road), 4 lanes, and 50 meter message range is also feasible since it has the same interferer number. VII. DISCUSSIONS Our evaluation exercise shows DSRC safety messaging on the control channel may be feasible. Messages generated every 200 msec is a good rate, since a driver reaction time of 0.7 seconds or higher in near-miss situations means, an on-board safety system relying on communicated messages may be able to recognize the situation faster, thereby providing timely assistance to the driver. Moreover, the interpretation of most safety information will require the sender to include its position in the message. Most likely, the position information Fig. 12. Probability of message failure bursts vs. repetition number: AFR-CS Protocol in the nominal setting Message Generation Interval 250 200 150 100 50 Feasible Infeasible 100 Bytes 250 Bytes 400 Bytes 0 0 20 40 60 80 100 120 140 160 180 200 Interferer Number Fig. 13. Feasibility Regions for Probability of Reception Failure < 0.01 and CBT < 50% will be derived from GPS. Most commercial off-theshelf GPS does not update faster than 5 Hz. The 250 byte message size is also adequate. If a highway reaches its capacity flow between 50 to 55 mph, at these speeds almost all light-duty passenger vehicles are able to easily stop within 150 meters. Thus this is an adequate value for the message range. Safety system designers have a reasonable chance of designing safety systems within these offered traffic limits. At these offered traffic levels the network may deliver a PRF as low as 1/100. However, the probability of successive losses is much lower. Thus if the vehicle does not learn of the emergency in 200 msec it will learn in 400 msec. Moreover, our evaluation is conducted assuming all vehicles transmit all the time. If commu-

nication becomes a general way of learning the vehicle neighborhood then this will be the case, i.e., each vehicle will regularly transmit its position, velocity, turn signal status, etc., for the benefit of others. Our analysis is in this case because it represents the largest offered traffic. Therefore, we think of our PRF s as a worst-case. This evaluation exercise also provides some design insights. Our results on the optimal data rate at different vehicular traffic densities, indicate the need for adaptive modulation control. All message ranges may not be feasible at all vehicular traffic densities. This indicates the need for power adaptation. This adaptation should be with respect to the flows of the vehicles. The repetition MAC is simple, easily added to 802.11a, works without infrastructure, and should deliver adequate performance on most rural and many urban or inter-urban roads. In some urban areas with high density traffic, e.g. 8-10 lanes, the PRF s may be too high. When there are many non-safety applications in the control channel such as service announcements, the CBT due to the safety traffic may be too high. In these settings, it may be best to use roadside radios to coordinate vehicle transmissions for more efficient use of the control channel. Vehicles would then follow the ad-hoc protocol till they detect a controlling roadside radio. Thereafter they would communicate according to the roadside control protocol and switch back to ad-hoc operation on not hearing the roadside radio again. This is an idea currently under investigation. ACKNOWLEDGMENT This work was supported by California PATH projects TO4224 and TO4210, and partially by a gift from Daimler-Chrysler Research and Technology North America, Inc. The authors thank Mr. Daniel Jiang of Daimler Chrysler RTNA and Dr. Hariharan Krishnan of General Motors Research and Development for valuable discussions. We thank Mr. Joel VanderWerf for help on vehicle traffic simulations, and Mr. Marc Torrent Moreno for help on NS-2 simulations. [6] B. Cash. IEEE briefing on 5.9 GHz concepts and band plan. Presentation in DSRC standardization monthly meeting, 2002. [7] Federal Communications Commission. FCC 99-305. FCC Report and Order, October 1999. [8] Federal Communications Commission. FCC 03-024. FCC Report and Order, Feburary 2004. [9] J. Garcia-Luna-Aceves and C. Fullmer. Floor aquisition multiple access (FAMA) in single channel wireless networks. ACM Mobile networks and applications, 4:157 174, 1999. [10] P. Gupta and P. Kumar. The capacity of wireless networks. IEEE Transactions on Information Theory, IT-46(2):388 404, March 2000. [11] P. Karn. MACA-a new channel access method for packet radio. ARRL/CRRL Amateur Radio 9th Computer Networking Conference, pages 134 140, 1990. [12] H. Krishnan and C. Kellum. Use of communication in vehicle safety application. Internal Report of General Motors Company, 2002. [13] W. C. Y. Lee. Mobile Communications Design Fundamentals. John Wiley & Sons, 2 edition, 1993. [14] California PATH. Shift: The hybrid system simulation programming language. http://www.path.berkeley.edu/shift/. [15] W. Pattra-Atikom, P. Krishnamurthy, and S. Banerjee. Distributed mechanisms for quality of serivce in wireless LAN. IEEE Wireless Communications, pages 26 34, June 2003. [16] J. Sobrinho and A. Krishnakumar. Quality-of-service in ad hoc carrier sense multiple access wireless networks. IEEE Journal on Selected Areas in Communications, 17(8):1353 1368, August 1999. [17] F. Tobagi and L. Kleinrock. Packet switching in radio channels: Part I- carrier sense multiple-access modes and their throughput/delay characteristics. IEEE Trans. Comm., COM- 23(12):1400 1416, December 1975. [18] J. VanderWerf, N. Kourjanskaia, S. Shladover, H. Krishnan, and M. Miller. Modeling the effects of driver control assistance systems on traffic. National Research Council Transportation Research Board 80th Annual Meeting, January 2001. [19] J. Walrand and P. Varaiya. High-performance Communications Networks. Morgan Kaufmann, 2 edition, 2000. [20] Y. Xiao. Enhanced DCF of IEEE 802.11e to support Qos. Proceedings of IEEE WCNC, pages 1291 1296, 2003. [21] Q. Xu, T. Mak, J. Ko, R. Sengupta, and D. Jiang. Ad hoc medium access control protocol design and analysis for vehicle safety communications. In preparation, 2004. [22] J. Zhu and S. Roy. MAC for Dedicated Short Range Communications in Intelligent Transporation System. IEEE Communications Magazine, pages 60 67, December 2003. REFERENCES [1] Dedicated Short Range Communications (DSRC) home. http://www.leearmstrong.com/dsrc/dsrchomeset.htm. [2] The netowrk simulator: NS-2. http://www.isi.edu/nsnam/ns. [3] N. Abramson. The throughput of packet broadcasting channels. IEEE Trans. Comm., COM-25:117 128, January 1977. [4] G. Anastasi, L. Lanzini, and E. Mingozzi. HIPERLAN/1 MAC protocol: stability and performance analysis. IEEE Journal on Selected Areas in Communications, 18(9):1787 1798, September 2000. [5] V. Bharghavan, A. Demers, S. Shanker, and L. Zhang. MACAW: A media access protocol for wireless LANs. ACM SIGCOMM 94, pages 212 225, August 1994.